A bank can cut 2,500 jobs. It cannot cut the work those jobs did

Two and a half thousand people leave. The work does not.

National Australia Bank is cutting about 2,500 roles, roughly six per cent of its staff, and the reason given is technological change. It is the most comfortable sentence a large bank can say, because it cannot be marked right or wrong. Nobody will stand up at the next results and ask whether the technology actually changed.

Compare it with a sentence that could be tested. “We have automated identity checks in account opening, so the team that did them by hand is no longer needed.” That is a claim with a date attached and a consequence you can measure six months later in error rates and reopened files. “Technological change” has neither. It is a category, not a cause.

This is not pedantry about wording. A restructuring justified by capability and a restructuring justified by cost are different decisions with different risks, and only one of them has to prove anything. Cost cutting is a legitimate thing for a bank to do and shareholders are entitled to want it. But a cost cut dressed as an automation dividend escapes the question that should follow it: what happens to the work?

Automation handles the middle. People handle the tail.

Software is very good at the median customer. The payment that goes through, the loan that fits the policy, the address change. What software handles badly is the case that does not fit — a hardship request, a disputed transaction, a deceased estate, a small business whose numbers are odd for a reason a system cannot see.

Those are the cases where a human decision is not a courtesy, it is the control. They are also, in most banks, where the experienced back-office and contact-centre staff sit. When headcount comes out faster than capability goes in, the cases do not disappear. They queue. And the cost of that queue does not land in the bank’s expense line, which is the point: it lands on the customer’s time and on the staff who are left holding the exceptions.

The hour the machines stop

It was reported this week that banks and websites across Europe went dark for hours when a Google Cloud data centre in London reported a service disruption. That is the argument for depth in one sentence. Running a bank in degraded mode — manual authorisations, phone queues, workarounds — is a skill held by people who have done it before. Thin staffing looks efficient right up to the hour it isn’t, and then it looks like something else entirely.

What the Reserve Bank just said out loud

There is a neat coincidence in the feeds. The Reserve Bank of Australia announced this month that its incoming chief risk officer, starting in January, is Rachael Phelan — most recently chief operating officer of NAB’s business and private bank, where the Bank’s own release says she led major transformation initiatives, and before that chief risk officer at NAB’s personal bank and uBank.

Read the release as a statement of what the central bank now thinks risk expertise consists of: governance, enterprise risk management, regulatory change and organisational transformation, listed together. Transformation is being treated as a risk discipline, not a management fashion. If that is the view at the institution that sits at the centre of the payments system, a major bank’s board cannot plausibly treat a six per cent reduction in staff as a purely commercial matter with no operational risk content attached.

The strongest case for the other side

Payroll is the largest controllable cost in a bank, and keeping people employed to do work that software does faster and more accurately is a subsidy that customers pay for in fees and margins. Banks have genuinely digitised: branch visits, cheque processing, manual reconciliation and paper verification have all shrunk, and the staffing that supported them should shrink too. Announced role reductions are also frequently met in part through attrition and redeployment rather than forced exits, so the human cost is usually smaller than the headline implies.

All of that is true, and none of it is an argument against evidence. If the automation is real, saying which functions it covers costs the bank nothing competitively — rivals know what onboarding automation looks like. The reluctance to be specific is not commercial sensitivity. It is that specificity creates a benchmark, and a benchmark can be missed.

Sequencing is the whole argument

The reason to insist on dates is that the two halves of this decision run on different clocks. Redundancies complete on a day. Systems go live late, partially, and with defects. A bank that takes the savings in this half and delivers the capability over the next three has not automated anything yet; it has borrowed service quality from its customers and promised to pay it back.